Canada’s defence minister Bill Blair said in September 2024 that the armed forces planned to become an “AI-enabled” organization by 2030 while keeping people responsible for decisions. His comments followed an international summit in Seoul on responsible military artificial intelligence.
The strategy was launched in March
The Department of National Defence and Canadian Armed Forces published their first joint AI strategy in March 2024. It described artificial intelligence as a capability that could support operations, administration and decision-making.
The document emphasized augmentation rather than replacement of human action.
Military AI covers many functions
Possible uses include equipment maintenance, logistics, cyber defence, intelligence analysis, training and processing large volumes of sensor data. Not every application controls a weapon or makes a battlefield decision.
Risk controls should be matched to the consequence of each system.
Human responsibility cannot be delegated
The Seoul blueprint endorsed by 61 states said responsibility and accountability cannot be transferred to machines. Appropriate human involvement remains necessary, particularly where force, detention or civilian safety is at stake.
A person clicking an approval button is not meaningful control unless that person understands the evidence and has time to intervene.
International law continues to apply
AI does not alter obligations under international humanitarian law, including distinction, proportionality and precautions in attack. A system’s novelty cannot excuse an unlawful outcome.
Legal review must cover intended use, foreseeable misuse and changes introduced through software updates.
Reliable data is a defence requirement
Models trained on incomplete, biased or manipulated data can make confident errors. Classified information, adversarial deception and rapidly changing conditions add risks not encountered in ordinary commercial tools.
Secure data governance and testing are therefore core operational safeguards.
Automation bias can weaken judgment
Personnel may defer to a system because its output appears technical or precise. Training should encourage challenge, require alternative assessments and record when commanders accept or reject automated recommendations.
Accountability depends on preserving a usable decision trail.
Procurement needs continuing oversight
A vendor demonstration cannot establish battlefield reliability. Contracts should permit independent testing, vulnerability reporting, audit access and safe withdrawal if performance changes.
Canada also has to manage supply-chain security without closing useful scrutiny behind claims of commercial secrecy.
International declarations were only a starting point
The Seoul document was a political blueprint, not an enforceable treaty. Shared principles can shape norms, but states still need domestic rules, transparent doctrine and mechanisms to investigate failures.
Countries with conflicting security interests may interpret “appropriate” human involvement differently.
The 2030 goal needed measurable safeguards
Success should not be counted only by the number of systems purchased. Defence officials could report high-level testing standards, workforce training, legal reviews and significant incidents without disclosing operational secrets.
Blair’s assurance addressed a central public concern, but durable trust required more than promising humans would remain present. Canada had to define who was responsible, what information they received and when a machine-assisted decision was too consequential to automate.
Parliamentary committees and independent oversight bodies could test those definitions over time. Rules also needed to cover experiments and intelligence tools, not only systems formally declared operational.
